RFE: repoclosure shouldn't fail on conditional dependencies

[bstinsonmhk]

We run repoclosure as part of the compose process to check that all deps are resolved. In a compose that contains packages with conditional dependencies we have a number of failures in the log files that look like this:

package: pyproject-srpm-macros-1.12.0-2.el10.noarch from CentOS-Stream-10-20240812.0-repoclosure-    AppStream.x86_64
  unresolved deps:
    (pyproject-rpm-macros = 1.12.0-2.el10 if pyproject-rpm-macros)

It would be great if there was a way to satisfy the conditional during the repoclosure run, but if not, we should make these warnings rather than failures.

[yselkowitz]

FWIW dnf5 repoclosure appears to have the same issue.

[jan-kolarik]

Hi Brian,

It seems there might be an issue with the arguments passed to repoclosure. I tested dnf repoclosure --pkg pyproject-srpm-macros in a Fedora 39 container, and it worked correctly—I got an empty output while I've checked the conditional dependency was present.

However, Fedora uses best=False, while on RHEL, it's set to best=True. This difference affects the behavior of the repoclosure plugin. When best=True, dependencies are resolved only against the latest versions of packages in the repositories, while with best=False, any package from the repositories can be used. This behavior is documented in the sources here, though the official user documentation isn't clear on this point. We should definitely work on improving it.

I think this issue only occurs when there are multiple versions of a package in the repositories. At this point, I don't see a better solution than using the --nobest option.

[m-blaha]

It does not work correctly for example in case the package after if does not exist. In such case the conditional requirement should be resolved as true, but it does not. (e.g. (NOT_EXIST > 3 if NOT_EXIST)).

[jan-kolarik]

@bstinsonmhk Bump. Would passing the --nobest work in your context or is there a different issue?